Im Wintersemester 2015/2016 findet im Rahmen des Informatik-Kolloquiums der folgende Vortrag statt:
Montag, 21. Dezember 2015, 16.15 Uhr, Turing-Hörsaal
Prof. Dr. Myra Spiliopoulou, Otto-von-Guericke-Universität Magdeburg
Leveraging Classification Quality through the Exploitation of Historical Unlabeled Medical Data
Epidemiological studies are used to derive diagnostic tests and to identify risk factors associated with a medical outcome. Longitudinal population-based studies can deliver insights about change in the population under study - change in the prevalence of the outcome, in the factors influencing it and in the interplay among those factors over time. However, whenever the protocol of a longitudinal study changes to include new assessments and new medical outcomes of interest, old recordings of the population participants become unusable because the new variables are missing from them. In this talk, I present methods for the exploitation of such old data for supervised learning, and I report results are from the analysis of a population-based cohort Study of Health in Pomerania — SHIP) on the outcome “hepatic steatosis” (fatty liver).